Ray-tracing is widely used for radio propagation modeling in indoor environments. Accurate simulations necessitate a high number of ray interactions with the environment, in terms of reflections, transmissions and diffractions, as well as the launching of a large number of rays. A machine learning approach can instead utilize fewer ray interactions and a smaller number of rays launched, to produce an accelerated output. In this paper, a generative adversarial network is trained by low resolution 2-dimensional maps of received signal strength in an office environment. Then, it generates higher resolution RSS maps, thereby significantly reducing the required simulation time. Apart from resolution compensation and solver acceleration, it can also accurately generalize to other frequencies and receiver points.